Dual-Arm Composition

DualArmRobotCfg builds a bimanual robot from a single-arm base config that already follows the standard "arm" convention. In the current EmbodiChain setup, the dual-arm examples are based on the UR family and are assembled onto a shared synthetic base_link through URDFCfg.

The images below show the three documented dual-arm layouts currently used for this robot family: a separated side-by-side arrangement, a wide mirrored forward-facing arrangement, and a tighter mirrored arrangement for close workspace overlap.

Side-by-Side

Wide Mirrored Layout

Close Mirrored Layout

Key Features

• Generic dual-arm builder from any single-arm RobotCfg that exposes one arm URDF component, one control_parts["arm"], and one solver_cfg["arm"].

• Image-backed mount presets covering separated, inward-facing, and mirrored-rz layouts.

• Automatic left/right derivation for URDF components, control parts, solver configs, and drive properties.

• Config-driven construction through DualArmRobotCfg.from_dict(...) with registry-based base robot lookup.

• Optional composite dual_arm part for commanding both manipulators together.

• Round-trip support through to_dict() / from_dict() on the generated config.

Visual Layouts

• Side-by-side keeps both bases offset by ±separation/2 along Y with the same orientation.

• Wide mirrored layout spreads the arms apart while rotating them symmetrically so they open away from each other.

• Close mirrored layout rotates the arms symmetrically toward a shared center workspace for overlapping reach.

Usage

from embodichain.lab.sim import SimulationManager, SimulationManagerCfg
from embodichain.lab.sim.robots import DualArmRobotCfg

sim = SimulationManager(SimulationManagerCfg(headless=True, num_envs=4))

cfg = DualArmRobotCfg.from_dict(
    {
        "base_robot": "ur5",
        "mount": {
            "preset": "mirrored_rz",
            "separation": 0.6,
            "rz": 0.7853981633974483,
        },
    }
)
robot = sim.add_robot(cfg=cfg)

Mount Presets

Preset	Layout
side_by_side	Left arm at +separation/2 in Y, right arm at -separation/2 in Y, same orientation on both sides.
facing_inward	Same ±separation/2 offsets, with yaw +pi/2 on the left arm and -pi/2 on the right arm.
mirrored_rz	Same ±separation/2 offsets, with yaw +rz on the left arm and -rz on the right arm.

The mount configuration also supports paired per-arm overrides:

cfg = DualArmRobotCfg.from_dict(
    {
        "base_robot": "ur5",
        "mount": {
            "preset": "side_by_side",
            "separation": 0.6,
            "left": {"xyz": [0.0, 0.35, 0.0], "rpy": [0.0, 0.0, 0.0]},
            "right": {"xyz": [0.1, -0.35, 0.0], "rpy": [0.0, 0.0, 0.0]},
        },
    }
)

Configuration Parameters

Parameter	Description
base_robot	Registry key such as "ur3" through "ur10e", or {"type": ..., "init": {...}} for extra base config overrides.
mount	Mount dict consumed by resolve_mounts, including preset, separation, optional rz, and optional paired left / right overrides.
arm_part	Base robot manipulator control part name. Default: "arm".
dual_part	Whether to emit the concatenated dual_arm control part. Default: True.

Derived Control Parts

For a UR5 base robot with the default preserved joint casing, DualArmRobotCfg produces:

Part	Joints
left_arm	left_joint1 to left_joint6
right_arm	right_joint1 to right_joint6
dual_arm	concatenation of the 12 joints above

For other base robots, the builder preserves the source URDF naming policy and applies the left_ / right_ prefixes consistently with the assembled URDF.

Programmatic Build Path

from embodichain.lab.sim.robots import URRobotCfg, build_dual_arm_cfg, resolve_mounts

base = URRobotCfg.from_dict({"robot_type": "ur5"})
mounts = resolve_mounts({"preset": "facing_inward", "separation": 0.6})
cfg = build_dual_arm_cfg(base, mounts, dual_part=False)

Adding a New Base Robot

A single-arm robot becomes dual-arm-ready by:

1. following the existing "arm" convention in its single-arm cfg,

2. exposing control_parts["arm"] and solver_cfg["arm"],

3. adding one registry entry in embodichain/lab/sim/robots/dual_arm.py.

That keeps the dual-arm path generic: no extra mixin or robot-specific dual-arm class is required.